drm/i915/fbc: split intel_fbc_update into pre and post update
authorPaulo Zanoni <paulo.r.zanoni@intel.com>
Tue, 19 Jan 2016 13:35:43 +0000 (11:35 -0200)
committerPaulo Zanoni <paulo.r.zanoni@intel.com>
Fri, 29 Jan 2016 20:11:48 +0000 (18:11 -0200)
So now pre_update will be responsible for unconditionally deactivating
FBC and updating the state cache, while post_update will be
responsible for checking if it can be enabled, then enabling it.

This is one more step into proper locking.

Notice that intel_fbc_flush now calls post_update directly. The FBC
flush can only happen for drawing operations - since we explicitly
ignore the flips -, so the FBC state is not expected to have changed
at this point. With this we can just run post_update, which will make
sure we won't deactivate+reactivate FBC as would be the case now if we
called pre_update + post_update.

Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1453210558-7875-11-git-send-email-paulo.r.zanoni@intel.com
drivers/gpu/drm/i915/intel_fbc.c

index 49d445a1854e67f1feec9e53f5567792d0fed8e2..61523cd95ac1a8b01153df348308f10066dceb25 100644 (file)
@@ -896,24 +896,16 @@ static bool intel_fbc_reg_params_equal(struct intel_fbc_reg_params *params1,
        return memcmp(params1, params2, sizeof(*params1)) == 0;
 }
 
-/**
- * __intel_fbc_update - activate/deactivate FBC as needed, unlocked
- * @crtc: the CRTC that triggered the update
- *
- * This function completely reevaluates the status of FBC, then activates,
- * deactivates or maintains it on the same state.
- */
-static void __intel_fbc_update(struct intel_crtc *crtc)
+static void intel_fbc_pre_update(struct intel_crtc *crtc)
 {
        struct drm_i915_private *dev_priv = crtc->base.dev->dev_private;
        struct intel_fbc *fbc = &dev_priv->fbc;
-       struct intel_fbc_reg_params old_params;
 
        WARN_ON(!mutex_is_locked(&fbc->lock));
 
        if (!multiple_pipes_ok(dev_priv)) {
                set_no_fbc_reason(dev_priv, "more than one pipe active");
-               goto out_disable;
+               goto deactivate;
        }
 
        if (!fbc->enabled || fbc->crtc != crtc)
@@ -921,8 +913,25 @@ static void __intel_fbc_update(struct intel_crtc *crtc)
 
        intel_fbc_update_state_cache(crtc);
 
-       if (!intel_fbc_can_activate(crtc))
-               goto out_disable;
+deactivate:
+       __intel_fbc_deactivate(dev_priv);
+}
+
+static void intel_fbc_post_update(struct intel_crtc *crtc)
+{
+       struct drm_i915_private *dev_priv = crtc->base.dev->dev_private;
+       struct intel_fbc *fbc = &dev_priv->fbc;
+       struct intel_fbc_reg_params old_params;
+
+       WARN_ON(!mutex_is_locked(&fbc->lock));
+
+       if (!fbc->enabled || fbc->crtc != crtc)
+               return;
+
+       if (!intel_fbc_can_activate(crtc)) {
+               WARN_ON(fbc->active);
+               return;
+       }
 
        old_params = fbc->params;
        intel_fbc_get_reg_params(crtc, &fbc->params);
@@ -936,44 +945,9 @@ static void __intel_fbc_update(struct intel_crtc *crtc)
            intel_fbc_reg_params_equal(&old_params, &fbc->params))
                return;
 
-       if (intel_fbc_is_active(dev_priv)) {
-               /* We update FBC along two paths, after changing fb/crtc
-                * configuration (modeswitching) and after page-flipping
-                * finishes. For the latter, we know that not only did
-                * we disable the FBC at the start of the page-flip
-                * sequence, but also more than one vblank has passed.
-                *
-                * For the former case of modeswitching, it is possible
-                * to switch between two FBC valid configurations
-                * instantaneously so we do need to disable the FBC
-                * before we can modify its control registers. We also
-                * have to wait for the next vblank for that to take
-                * effect. However, since we delay enabling FBC we can
-                * assume that a vblank has passed since disabling and
-                * that we can safely alter the registers in the deferred
-                * callback.
-                *
-                * In the scenario that we go from a valid to invalid
-                * and then back to valid FBC configuration we have
-                * no strict enforcement that a vblank occurred since
-                * disabling the FBC. However, along all current pipe
-                * disabling paths we do need to wait for a vblank at
-                * some point. And we wait before enabling FBC anyway.
-                */
-               DRM_DEBUG_KMS("deactivating FBC for update\n");
-               __intel_fbc_deactivate(dev_priv);
-       }
-
+       __intel_fbc_deactivate(dev_priv);
        intel_fbc_schedule_activation(crtc);
-       fbc->no_fbc_reason = "FBC enabled (not necessarily active)";
-       return;
-
-out_disable:
-       /* Multiple disables should be harmless */
-       if (intel_fbc_is_active(dev_priv)) {
-               DRM_DEBUG_KMS("unsupported config, deactivating FBC\n");
-               __intel_fbc_deactivate(dev_priv);
-       }
+       fbc->no_fbc_reason = "FBC enabled (active or scheduled)";
 }
 
 /*
@@ -991,7 +965,8 @@ void intel_fbc_update(struct intel_crtc *crtc)
                return;
 
        mutex_lock(&fbc->lock);
-       __intel_fbc_update(crtc);
+       intel_fbc_pre_update(crtc);
+       intel_fbc_post_update(crtc);
        mutex_unlock(&fbc->lock);
 }
 
@@ -1045,7 +1020,7 @@ void intel_fbc_flush(struct drm_i915_private *dev_priv,
                if (fbc->active)
                        intel_fbc_recompress(dev_priv);
                else
-                       __intel_fbc_update(fbc->crtc);
+                       intel_fbc_post_update(fbc->crtc);
        }
 
        mutex_unlock(&fbc->lock);